
Which direction you are facing today?
Concept LoRas that removes focus on characters and allows them to be placed on sides of pictures more reliably. Because of this models can generate better objects and landscapes beside character.
Most visible on landscape 768x512 resolution, you can even go for 1024x512 and still receive very good results.
It will work on square or portrait but won't be this much noticeable
I tried to mix both left and right side but I wasn't able to trigger specific side, I was just getting random side. Hence why two versions, at least it's consistent.
Works almost every time on Anime/NAI related models, but I'm not sure about realistic checkpoints because my dummy brain forgot to add some realistic dataset >.<
Weight strength : 0.8-1.0
Trigger words :
Trigger is not needed for this one, just put your LoRA in and it should work. (after some testing I noticed that triggers don't do really much... which is weird but might just be issue with unique concept, dunno)
But if it will help someone, here :
-
lside, left side
-
rside, right side
描述:
Just put LoRA in and should be good to go.
Keep weight above 0.8
训练词语:
名称: left side position.safetensors
大小 (KB): 18571
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success